Mercurial > libavcodec.hg
annotate flashsvenc.c @ 4735:8903c1d6db18 libavcodec
simplify
author | michael |
---|---|
date | Mon, 26 Mar 2007 00:44:42 +0000 |
parents | ff9749708137 |
children | b3ee9a1526b0 |
rev | line source |
---|---|
4374 | 1 /* |
2 * Flash Screen Video encoder | |
3 * Copyright (C) 2004 Alex Beregszaszi | |
4 * Copyright (C) 2006 Benjamin Larsson | |
5 * | |
6 * This file is part of FFmpeg. | |
7 * | |
8 * FFmpeg is free software; you can redistribute it and/or | |
9 * modify it under the terms of the GNU Lesser General Public | |
10 * License as published by the Free Software Foundation; either | |
11 * version 2.1 of the License, or (at your option) any later version. | |
12 * | |
13 * FFmpeg is distributed in the hope that it will be useful, | |
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 * Lesser General Public License for more details. | |
17 * | |
18 * You should have received a copy of the GNU Lesser General Public | |
19 * License along with FFmpeg; if not, write to the Free Software | |
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
21 */ | |
22 | |
23 /* Encoding development sponsored by http://fh-campuswien.ac.at */ | |
24 | |
25 /** | |
26 * @file flashsvenc.c | |
27 * Flash Screen Video encoder | |
28 * @author Alex Beregszaszi | |
29 * @author Benjamin Larsson | |
30 */ | |
31 | |
32 /* Bitstream description | |
33 * The picture is divided into blocks that are zlib-compressed. | |
34 * | |
35 * The decoder is fed complete frames, the frameheader contains: | |
36 * 4bits of block width | |
37 * 12bits of frame width | |
38 * 4bits of block height | |
39 * 12bits of frame height | |
40 * | |
41 * Directly after the header are the compressed blocks. The blocks | |
42 * have their compressed size represented with 16bits in the beginig. | |
43 * If the size = 0 then the block is unchanged from the previous frame. | |
44 * All blocks are decompressed until the buffer is consumed. | |
45 * | |
46 * Encoding ideas, a basic encoder would just use a fixed block size. | |
47 * Block sizes can be multipels of 16, from 16 to 256. The blocks don't | |
48 * have to be quadratic. A brute force search with a set of different | |
49 * block sizes should give a better result than to just use a fixed size. | |
50 */ | |
51 | |
52 /* TODO: | |
53 * Don't reencode the frame in brute force mode if the frame is a dupe. Speed up. | |
54 * Make the difference check faster. | |
55 */ | |
56 | |
57 #include <stdio.h> | |
58 #include <stdlib.h> | |
59 #include <zlib.h> | |
60 | |
61 #include "common.h" | |
62 #include "avcodec.h" | |
63 #include "bitstream.h" | |
64 #include "bytestream.h" | |
65 | |
66 | |
67 typedef struct FlashSVContext { | |
68 AVCodecContext *avctx; | |
69 uint8_t *previous_frame; | |
70 AVFrame frame; | |
71 int image_width, image_height; | |
72 int block_width, block_height; | |
73 uint8_t* tmpblock; | |
74 uint8_t* encbuffer; | |
75 int block_size; | |
76 z_stream zstream; | |
4651
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
77 int last_key_frame; |
4374 | 78 } FlashSVContext; |
79 | |
80 static int copy_region_enc(uint8_t *sptr, uint8_t *dptr, | |
81 int dx, int dy, int h, int w, int stride, uint8_t *pfptr) { | |
82 int i,j; | |
83 uint8_t *nsptr; | |
84 uint8_t *npfptr; | |
85 int diff = 0; | |
86 | |
87 for (i = dx+h; i > dx; i--) { | |
88 nsptr = sptr+(i*stride)+dy*3; | |
89 npfptr = pfptr+(i*stride)+dy*3; | |
90 for (j=0 ; j<w*3 ; j++) { | |
91 diff |=npfptr[j]^nsptr[j]; | |
92 dptr[j] = nsptr[j]; | |
93 } | |
94 dptr += w*3; | |
95 } | |
96 if (diff) | |
97 return 1; | |
98 return 0; | |
99 } | |
100 | |
101 static int flashsv_encode_init(AVCodecContext *avctx) | |
102 { | |
103 FlashSVContext *s = (FlashSVContext *)avctx->priv_data; | |
104 | |
105 s->avctx = avctx; | |
106 | |
107 if ((avctx->width > 4095) || (avctx->height > 4095)) { | |
108 av_log(avctx, AV_LOG_ERROR, "Input dimensions too large, input must be max 4096x4096 !\n"); | |
109 return -1; | |
110 } | |
111 | |
112 if (avcodec_check_dimensions(avctx, avctx->width, avctx->height) < 0) { | |
113 return -1; | |
114 } | |
115 | |
116 // Needed if zlib unused or init aborted before deflateInit | |
117 memset(&(s->zstream), 0, sizeof(z_stream)); | |
118 /* | |
119 s->zstream.zalloc = NULL; //av_malloc; | |
120 s->zstream.zfree = NULL; //av_free; | |
121 s->zstream.opaque = NULL; | |
122 zret = deflateInit(&(s->zstream), 9); | |
123 if (zret != Z_OK) { | |
124 av_log(avctx, AV_LOG_ERROR, "Inflate init error: %d\n", zret); | |
125 return -1; | |
126 } | |
127 */ | |
128 | |
4651
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
129 s->last_key_frame=0; |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
130 |
4374 | 131 s->image_width = avctx->width; |
132 s->image_height = avctx->height; | |
133 | |
134 s->tmpblock = av_mallocz(3*256*256); | |
135 s->encbuffer = av_mallocz(s->image_width*s->image_height*3); | |
136 | |
137 if (!s->tmpblock || !s->encbuffer) { | |
138 av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); | |
139 return -1; | |
140 } | |
141 | |
142 return 0; | |
143 } | |
144 | |
145 | |
146 static int encode_bitstream(FlashSVContext *s, AVFrame *p, uint8_t *buf, int buf_size, | |
147 int block_width, int block_height, uint8_t *previous_frame, int* I_frame) { | |
148 | |
149 PutBitContext pb; | |
150 int h_blocks, v_blocks, h_part, v_part, i, j; | |
151 int buf_pos, res; | |
152 int pred_blocks = 0; | |
153 | |
154 init_put_bits(&pb, buf, buf_size*8); | |
155 | |
156 put_bits(&pb, 4, (block_width/16)-1); | |
157 put_bits(&pb, 12, s->image_width); | |
158 put_bits(&pb, 4, (block_height/16)-1); | |
159 put_bits(&pb, 12, s->image_height); | |
160 flush_put_bits(&pb); | |
161 buf_pos=4; | |
162 | |
163 h_blocks = s->image_width / block_width; | |
164 h_part = s->image_width % block_width; | |
165 v_blocks = s->image_height / block_height; | |
166 v_part = s->image_height % block_height; | |
167 | |
168 /* loop over all block columns */ | |
169 for (j = 0; j < v_blocks + (v_part?1:0); j++) | |
170 { | |
171 | |
172 int hp = j*block_height; // horiz position in frame | |
173 int hs = (j<v_blocks)?block_height:v_part; // size of block | |
174 | |
175 /* loop over all block rows */ | |
176 for (i = 0; i < h_blocks + (h_part?1:0); i++) | |
177 { | |
178 int wp = i*block_width; // vert position in frame | |
179 int ws = (i<h_blocks)?block_width:h_part; // size of block | |
180 int ret=Z_OK; | |
181 uint8_t *ptr; | |
182 | |
183 ptr = buf+buf_pos; | |
184 | |
185 //copy the block to the temp buffer before compression (if it differs from the previous frame's block) | |
186 res = copy_region_enc(p->data[0], s->tmpblock, s->image_height-(hp+hs+1), wp, hs, ws, p->linesize[0], previous_frame); | |
187 | |
188 if (res || *I_frame) { | |
189 unsigned long zsize; | |
190 zsize = 3*block_width*block_height; | |
191 ret = compress2(ptr+2, &zsize, s->tmpblock, 3*ws*hs, 9); | |
192 | |
193 | |
194 //ret = deflateReset(&(s->zstream)); | |
195 if (ret != Z_OK) | |
196 av_log(s->avctx, AV_LOG_ERROR, "error while compressing block %dx%d\n", i, j); | |
197 /* | |
198 s->zstream.next_in = s->tmpblock; | |
199 s->zstream.avail_in = 3*ws*hs; | |
200 s->zstream.total_in = 0; | |
201 | |
202 s->zstream.next_out = ptr+2; | |
203 s->zstream.avail_out = buf_size-buf_pos-2; | |
204 s->zstream.total_out = 0; | |
205 | |
206 ret = deflate(&(s->zstream), Z_FINISH); | |
207 if ((ret != Z_OK) && (ret != Z_STREAM_END)) | |
208 av_log(s->avctx, AV_LOG_ERROR, "error while compressing block %dx%d\n", i, j); | |
209 | |
210 size = s->zstream.total_out; | |
211 //av_log(avctx, AV_LOG_INFO, "compressed blocks: %d\n", size); | |
212 */ | |
213 bytestream_put_be16(&ptr,(unsigned int)zsize); | |
4377
75942fdfd1a1
Add flashsv encoder to changelog and fix the encoder so it actually works.
banan
parents:
4374
diff
changeset
|
214 buf_pos += zsize+2; |
4374 | 215 //av_log(avctx, AV_LOG_ERROR, "buf_pos = %d\n", buf_pos); |
216 } else { | |
217 pred_blocks++; | |
218 bytestream_put_be16(&ptr,0); | |
4377
75942fdfd1a1
Add flashsv encoder to changelog and fix the encoder so it actually works.
banan
parents:
4374
diff
changeset
|
219 buf_pos += 2; |
4374 | 220 } |
221 } | |
222 } | |
223 | |
224 if (pred_blocks) | |
225 *I_frame = 0; | |
226 else | |
227 *I_frame = 1; | |
228 | |
229 return buf_pos; | |
230 } | |
231 | |
232 | |
233 static int flashsv_encode_frame(AVCodecContext *avctx, uint8_t *buf, int buf_size, void *data) | |
234 { | |
235 FlashSVContext * const s = (FlashSVContext *)avctx->priv_data; | |
236 AVFrame *pict = data; | |
237 AVFrame * const p = &s->frame; | |
238 int res; | |
239 int I_frame = 0; | |
240 int opt_w, opt_h; | |
241 | |
242 *p = *pict; | |
243 | |
4651
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
244 /* First frame needs to be a keyframe */ |
4644 | 245 if (avctx->frame_number == 0) { |
4406
548b2c0a2a51
Don't copy more then needed, bugfix by Gus Scheidt ellocogato at gmail dot com and Michel Bardiaux mbardiaux at mediaxim dot be.
banan
parents:
4378
diff
changeset
|
246 s->previous_frame = av_mallocz(p->linesize[0]*s->image_height); |
4374 | 247 if (!s->previous_frame) { |
248 av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); | |
249 return -1; | |
250 } | |
251 I_frame = 1; | |
252 } | |
253 | |
4651
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
254 /* Check the placement of keyframes */ |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
255 if (avctx->gop_size > 0) { |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
256 if (avctx->frame_number >= s->last_key_frame + avctx->gop_size) { |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
257 I_frame = 1; |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
258 } |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
259 } |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
260 |
4374 | 261 #if 0 |
262 int w, h; | |
263 int optim_sizes[16][16]; | |
264 int smallest_size; | |
265 //Try all possible combinations and store the encoded frame sizes | |
266 for (w=1 ; w<17 ; w++) { | |
267 for (h=1 ; h<17 ; h++) { | |
268 optim_sizes[w-1][h-1] = encode_bitstream(s, p, s->encbuffer, s->image_width*s->image_height*4, w*16, h*16, s->previous_frame); | |
269 //av_log(avctx, AV_LOG_ERROR, "[%d][%d]size = %d\n",w,h,optim_sizes[w-1][h-1]); | |
270 } | |
271 } | |
272 | |
273 //Search for the smallest framesize and encode the frame with those parameters | |
274 smallest_size=optim_sizes[0][0]; | |
275 opt_w = 0; | |
276 opt_h = 0; | |
277 for (w=0 ; w<16 ; w++) { | |
278 for (h=0 ; h<16 ; h++) { | |
279 if (optim_sizes[w][h] < smallest_size) { | |
280 smallest_size = optim_sizes[w][h]; | |
281 opt_w = w; | |
282 opt_h = h; | |
283 } | |
284 } | |
285 } | |
286 res = encode_bitstream(s, p, buf, buf_size, (opt_w+1)*16, (opt_h+1)*16, s->previous_frame); | |
287 av_log(avctx, AV_LOG_ERROR, "[%d][%d]optimal size = %d, res = %d|\n", opt_w, opt_h, smallest_size, res); | |
288 | |
289 if (buf_size < res) | |
290 av_log(avctx, AV_LOG_ERROR, "buf_size %d < res %d\n", buf_size, res); | |
291 | |
292 #else | |
293 opt_w=1; | |
294 opt_h=1; | |
295 | |
296 if (buf_size < s->image_width*s->image_height*3) { | |
297 //Conservative upper bound check for compressed data | |
298 av_log(avctx, AV_LOG_ERROR, "buf_size %d < %d\n", buf_size, s->image_width*s->image_height*3); | |
299 return -1; | |
300 } | |
301 | |
302 res = encode_bitstream(s, p, buf, buf_size, opt_w*16, opt_h*16, s->previous_frame, &I_frame); | |
303 #endif | |
304 //save the current frame | |
4406
548b2c0a2a51
Don't copy more then needed, bugfix by Gus Scheidt ellocogato at gmail dot com and Michel Bardiaux mbardiaux at mediaxim dot be.
banan
parents:
4378
diff
changeset
|
305 memcpy(s->previous_frame, p->data[0], s->image_height*p->linesize[0]); |
4374 | 306 |
307 //mark the frame type so the muxer can mux it correctly | |
308 if (I_frame) { | |
309 p->pict_type = FF_I_TYPE; | |
310 p->key_frame = 1; | |
4651
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
311 s->last_key_frame = avctx->frame_number; |
ff9749708137
Respect the gop size (-g) for marking I frames. Use -g 0 gives the old behaviour.
banan
parents:
4644
diff
changeset
|
312 av_log(avctx, AV_LOG_DEBUG, "Inserting key frame at frame %d\n",avctx->frame_number); |
4374 | 313 } else { |
314 p->pict_type = FF_P_TYPE; | |
315 p->key_frame = 0; | |
316 } | |
317 | |
318 avctx->coded_frame = p; | |
319 | |
320 return res; | |
321 } | |
322 | |
323 static int flashsv_encode_end(AVCodecContext *avctx) | |
324 { | |
325 FlashSVContext *s = (FlashSVContext *)avctx->priv_data; | |
326 | |
327 deflateEnd(&(s->zstream)); | |
328 | |
329 av_free(s->encbuffer); | |
330 av_free(s->previous_frame); | |
331 av_free(s->tmpblock); | |
332 | |
333 return 0; | |
334 } | |
335 | |
336 AVCodec flashsv_encoder = { | |
337 "flashsv", | |
338 CODEC_TYPE_VIDEO, | |
339 CODEC_ID_FLASHSV, | |
340 sizeof(FlashSVContext), | |
341 flashsv_encode_init, | |
342 flashsv_encode_frame, | |
343 flashsv_encode_end, | |
344 .pix_fmts = (enum PixelFormat[]){PIX_FMT_BGR24, -1}, | |
345 }; | |
346 |